Research And Implementation Of Noise Intensity Estimation Based On Hyperspectral Images

In the 1980 s,with the development of remote sensing technology,People can get the spatial and spectral information of earth objects at the same moment,accomplishing the integration of atlas and solving many unsolvable problems by panchromatic or multispectral transmission.With the rapid development of spectral imaging technology,several satellites equipped with hyperspectral imagers have been launched and put into service at home and abroad.The data obtained from these sensors will further support the development and application of HSI.However,the intensity of noise in the data is one of the most important evaluation index to the quality of images obtained by these sensors.Therefore,accurately assessing the signal-to-noise ratio in the image.Now the urgently needed problem which to analyze the influence of noise intensity on hyperspectral image applications to be solved.In order to accurately estimate the noise intensity of hyperspectral images,this article proposes a noise estimation intensity algorithm based on the doubled reference band linear prediction.Using inter-spectral correlation can improve the accuracy of noise assessment.Aiming at the fact that the existing algorithms fail to take full the strong spectral correlation of hyperspectral images,a dual reference band linear prediction model is proposed.First the reference band is selected,reconstruct the selected reference band and establish a linear prediction model of dual reference bands in line with the refactoring results.Then use the obtained vector to iteratively correct the predicted value until the predicted value no longer changes,the result obtained at this time is the noise estimation value of the hyperspectral image.The noise estimation in this article uses simulated and real hyperspectral data respectively.From the results,our algorithm has better merit than others in accuracy and robustness of noise estimation.The fault of noise is about 7% and with additional noise compared to the conventional method.it can still provide stable results.This paper also explores the Influence of noise on hyperspectral image classification.Experiments on the effect of noise on hyperspectral image classification,conducted on real datasets Indian Pines and Cuprite data.Then the results of the above experiments are analyzed.From the results obtained show that the classification accuracy of HSI slowly decreases when added different intensities and types of noise.When added different types of noise to Cuprite data,The classification accuracy also fluctuated to varying degrees.and the error is close to 10%.This shows that different signal-to-noise ratios have a large impact on classification accuracy.Therefore,it is necessary to carry out specific quantitative analysis on the analysis and use of noise on hyperspectral images.